Duck Creek Buys Send: The Build-vs-Buy Shift in Agentic Underwriting

Duck Creek acquired underwriting orchestration vendor Send Technology Solutions on July 7, 2026, folding in a platform that already carries $26 billion of gross written premium and claims up to 7x faster time-to-quote.

The layer being consolidated is not the scoring model. It is the workflow that decides which submissions a model ever sees, which means a carrier's appetite rules are moving into a vendor's configuration file.

Key Takeaways

• $26 billion of gross written premium already runs through Send's rules across commercial, specialty, MGA, delegated authority, reinsurance and London Market business, spanning more than 40 lines.
• 7x faster time-to-quote and a 65% shorter product launch cycle are both throughput measures, and two different mechanisms produce the first one.
• Ten weeks after launching its own orchestration workbench, Duck Creek bought a specialist instead of extending it.
• $310 million was EXL's price for iMerit in June 2026, $170 million upfront plus $140 million of earnouts, buying the training and annotation layer rather than orchestration.
• The NAIC AI Model Bulletin, adopted December 2023, requires written standards for third-party AI, vendor due diligence, and contractual audit rights, and is in force in close to half of US states.

What Send's Orchestration Engine Automates

Send is not a score or a rating engine. It sits between the arrival of a submission and the underwriter's decision, ingesting submissions from email, documents and broker channels, structuring unstructured data, flagging missing fields, and coordinating the workflow through risk assessment, pricing, approvals, quote, bind and post-bind.

It was built for the segments where submission complexity is the throughput constraint: commercial, specialty, MGA, delegated authority, reinsurance and London Market, where multi-section programs and subscription placements make manual data assembly the drag.

The underwriter has not left the workflow, but their position in it moved. Configurable rules decide which submissions auto-quote within appetite, which escalate for manual review, and which auto-decline before anyone opens the file. "Send was built to help insurers navigate increasingly complex underwriting decisions by orchestrating people, data, AI, and workflows into a single underwriting experience," said Andy Moss, Send's chief executive and co-founder, who joins Duck Creek as general manager of underwriting.

Duck Creek had already built this layer. It launched an insurance-native Agentic AI Platform on April 28, 2026 with an Agentic Orchestration component and an Agentic Underwriting Workbench. Ten weeks later it bought rules already proven against $26 billion of live premium instead.

7x Faster Measures Throughput, Not Selection

A system quotes faster when it processes more submissions automatically inside a defined rule set. It also quotes faster when it declines or routes away more submissions that fall outside that set before a human sees them. Both produce the same headline, and only the first is an efficiency gain rather than a change in what is being counted.

That is the adverse selection question the metric cannot answer. If the triage rules systematically over-quote or under-price a class, territory or risk profile against the carrier's actual experience, nothing fails at deployment. It emerges eighteen to twenty-four months later as unexplained loss ratio deterioration in exactly the segments the engine was quoting fastest, by which point a full accident year has been written on rules nobody outside the vendor re-tested.

The evidence that would settle it is narrow and specific: loss ratio by disposition, separating auto-quoted, escalated and auto-declined submissions, tied to the rule version in force when each policy was written. Without that split the 7x figure describes speed and says nothing about the risks that moved through fastest.

The comparison with the other recent deal sharpens what was bought. EXL agreed to acquire iMerit on June 24, 2026 for $310 million, $170 million upfront and $140 million in earnouts, picking up the Ango Hub annotation platform and the Scholars expert network. That is the training and evaluation layer, upstream of deployment. Send is downstream, where a model that already exists gets wired into a live decision.

Competitors have drawn the boundary differently. Guidewire runs an Agentic Framework letting carriers build or plug in agents on its cloud APIs, Sapiens has taken a partner-integration posture, and Socotra leaves orchestration to whatever a carrier builds. Convr announced the same day that it delivers intake, enrichment, classification, scoring and agentic decisioning across all three cores, describing itself as connective tissue rather than a replacement. Guidewire and Sapiens carriers still rent that layer. Duck Creek now owns it.

The Underwriting Manual Now Lives in a Vendor's Configuration

Orchestration engines do not route paperwork. They encode appetite thresholds and auto-quote conditions as executable rules, which means the working content of an underwriting manual increasingly sits as configuration inside third-party software. Send markets configurable workflows, rules, governance and compliance controls as the product. That configurability is also the exposure.

Triage and auto-decline logic that shapes which risks reach a pricing engine functions as underwriting criteria even where it is not a rating factor, and most states require an insurer to produce the criteria behind a filing on request. The NAIC Model Bulletin, adopted December 2023 and in force in close to half of US states, requires written standards for acquiring and using third-party AI, due diligence on vendor data and models, and contractual audit rights.

A carrier that cannot produce the current rule set, its version history and the loss experience validating it on the same timeline as its own manual has not met that standard simply because the vendor ships governance tooling. The Third-Party Data and Models Task Force, formed in 2024, is the body most likely to test the gap.

The commercial structure makes that harder to unwind than a scoring contract. A score is an API endpoint; an orchestration engine is thousands of triage and appetite rules in a proprietary format, and migrating means re-encoding them rather than repointing an integration. The 65% product launch figure describes a rules layer being reconfigured continuously, so every change is either a carrier request through the vendor or a vendor change the carrier has to re-validate, for the life of the contract. Carriers without mature in-house rules authoring have the strongest case for buying, and the least leverage once they have.